In this challenge, you will build models to predict the weather (specifically the temperature) at different locations around the globe. 

In this challenge, you will be exposed to the concept of covariate shift. The mismatch between training and deployment data adversely impacts ML models and is ubiquitous in real, industrial applications. 

To analyze the robustness of learned models to climate shifts, we use the Koppen climate classification which provides data to map latitudes and longitudes to one of five main climate types: Tropical, Dry, Mild Temperate, Snow and Polar. The dataset is augmented such that each sample has an associated climate type. 

The training data consists of ~2M measurements made from September 1st, 2018 till February 28th, 2019 for climate types Dry and Mild Temperate (note that the climate type is not used as a training feature).
The evaluation data will also include measurements for Tropical climate.